Research On Temproal Error Concealment Technology On H.264 SVC

As the Scalable Video Coding (SVC) extension of the H.264/AVC standard, H.264/AVC SVC incorporates efficiently the functionalities of scalability of temporal,spatial and quality as well as the good encoding performance.It supports transmitting and decoding partial bit streams to meet the constraint of different applications and transport environment. However, such a compressed video is very sensitive to data loss in error-prone channels(e.g., IP or wireless communication networks). Channel errors can not only damage current frame, but also propagate to subsequent frames, which can significantly degrade the image quality. Error concealment(EC), which doesn't impose any additional complexity on video encoder, provides a feasible solution for error handling by recovering the damaged areas, using spatial and temporal redundancy information of images and characteristics of human visual system. Therefore, it is a promising technique in theory and practice to design error concealment methods based on H.264/AVC SVC.In this thesis, firstly, we briefly outline the main techniques in the order of coding framework,temporal,spatial and quality scalability in H.264/AVC SVC,and then describe some error concealment technology applied in H.264. After detailed analysis of the existing SVC temporal concealment algorithm,this thesis proposes the following improvements:1, according to the characteristics of SVC temproal scalability and the decoder error concealment error propagation, proposed two-way TD concealment algorithm , making the choice of reference frame is more suited to lost frame; 2, proposed an improved motion vector estimation algorithm: expanding the candidate vector of missing one, and then improve the matching algorithm to estimate the best candidate vector; 3, according to the error conditions of base layer, adaptive select the BLSKIP or improved TD algorithm for the error concealment of lost enhancement layer. Finally, simulation results verify the improved algorithm. The experiment results show that the refined algorithm can aehieve 0.4-1.1 dB average PSNR gains for reeovered imagese compared with the original algorithm .